2. Inferring Threat from Scientific Collections
23
because one test happened to fare poorly on a given data set (Grimson et al. 1992;
Grimson 1993). This is particularly relevant in the analysis of small data sets,
conventionally with fewer than 30 observations, in which levels of accuracy may
be unsatisfactory or unknown for asymptotic methods that are usually applied to
sample sizes larger than 30. Paradoxically, it is easier to detect decline and
extinction in species that are well represented in a database than in species for
which there are few records (Solow 1993).
Other simple rules may be applied to identify taxa that may need protection or
management. For example, one may flag all species known from fewer than three
collections, or all species not collected for more than 20 years, or all species for
which collections may be enclosed by a convex polygon of less than 10 km
2
(IUCN 1994).
Extinctions are difficult to observe and they are a crude measure of impact
(Burgman et al. 1995). Conservation biology is more concerned with threatened
species, not those already lost. Observational data, particularly those in herbaria
and museums, hold valuable information that may be used to make inferences
about the status of species. There are two important points to consider in applying
the methods described above. The first is that they do not assume that the process
of collection has been fixed. Rather, they assume a stationary random process.
The second is that they do not replace conventional wisdom or the application of
more usual devices to establish conservation status. If they are informative or raise
questions in the minds of those who establish conservation priorities, then they
will have served their purpose. The application outlined here was successful
because several taxa were identified for survey that would not otherwise have
attracted priority attention. Explanations were found for most of the very unusual
patterns. However, for several taxa there was no plausible explanation, and further
surveys may be indicated. In this way, the methods may assist in the setting of
conservation priorities. The lesson is that the results of analysis from the equations
should not be interpreted in isolation. Rather, they should be used in conjunction
with a sound understanding of the day-to-day practices that affect collection rates.
Acknowledgments. The authors gratefully acknowledge Kirsten Howlett, Ken
Atkins, David Coates, and Terry Walshe for their thoughtful and valued comments
on this project. This work was supported by a small grant from the Australian
Research Council and by funding from the Australian Nature Conservation
Agency (Environment Australia) administered by Andrew Taplin.
Literature Cited
Bradley JN (1968) Distribution-free statistical tests. Prentice Hall, Englewood Cliffs, New
Jersey
Burgman MA, Grimson RC, Ferson S (1995) Inferring threat from scientific collections.
Conservation Biology 9:923–928
Grimson RC (1993) Disease clusters, exact distributions of maxima, and P-values. Statistics in Medicine 12:1773–1794
Grimson RC, Aldrich TE, Wanzer Drane J (1992) Clustering in sparse data and an analysis
of Rhabdomyosarcoma incidence. Statistics in Medicine 11:761–768
23
because one test happened to fare poorly on a given data set (Grimson et al. 1992;
Grimson 1993). This is particularly relevant in the analysis of small data sets,
conventionally with fewer than 30 observations, in which levels of accuracy may
be unsatisfactory or unknown for asymptotic methods that are usually applied to
sample sizes larger than 30. Paradoxically, it is easier to detect decline and
extinction in species that are well represented in a database than in species for
which there are few records (Solow 1993).
Other simple rules may be applied to identify taxa that may need protection or
management. For example, one may flag all species known from fewer than three
collections, or all species not collected for more than 20 years, or all species for
which collections may be enclosed by a convex polygon of less than 10 km
2
(IUCN 1994).
Extinctions are difficult to observe and they are a crude measure of impact
(Burgman et al. 1995). Conservation biology is more concerned with threatened
species, not those already lost. Observational data, particularly those in herbaria
and museums, hold valuable information that may be used to make inferences
about the status of species. There are two important points to consider in applying
the methods described above. The first is that they do not assume that the process
of collection has been fixed. Rather, they assume a stationary random process.
The second is that they do not replace conventional wisdom or the application of
more usual devices to establish conservation status. If they are informative or raise
questions in the minds of those who establish conservation priorities, then they
will have served their purpose. The application outlined here was successful
because several taxa were identified for survey that would not otherwise have
attracted priority attention. Explanations were found for most of the very unusual
patterns. However, for several taxa there was no plausible explanation, and further
surveys may be indicated. In this way, the methods may assist in the setting of
conservation priorities. The lesson is that the results of analysis from the equations
should not be interpreted in isolation. Rather, they should be used in conjunction
with a sound understanding of the day-to-day practices that affect collection rates.
Acknowledgments. The authors gratefully acknowledge Kirsten Howlett, Ken
Atkins, David Coates, and Terry Walshe for their thoughtful and valued comments
on this project. This work was supported by a small grant from the Australian
Research Council and by funding from the Australian Nature Conservation
Agency (Environment Australia) administered by Andrew Taplin.
Literature Cited
Bradley JN (1968) Distribution-free statistical tests. Prentice Hall, Englewood Cliffs, New
Jersey
Burgman MA, Grimson RC, Ferson S (1995) Inferring threat from scientific collections.
Conservation Biology 9:923–928
Grimson RC (1993) Disease clusters, exact distributions of maxima, and P-values. Statistics in Medicine 12:1773–1794
Grimson RC, Aldrich TE, Wanzer Drane J (1992) Clustering in sparse data and an analysis
of Rhabdomyosarcoma incidence. Statistics in Medicine 11:761–768